Versatile-Image-Compression
NOTE: A new pytorch implementation of iWave++ has been released at (https://gitlab.com/iWave/iwave), including training and testing code, and pre-trained models. This pytorch implementation is also used as the reference software for IEEE 1857.11 Standard for Neural Network-Based Image Coding.
This repo provides the official implementation of "End-to-End Optimized Versatile Image Compression With Wavelet-Like Transform".
Accepted by IEEE TPAMI.
Author: Haichuan Ma, Dong Liu, Ning Yan, Houqiang Li, Feng Wu
BibTeX
@article{ma2020end, title={End-to-End Optimized Versatile Image Compression With Wavelet-Like Transform}, author={Ma, Haichuan and Liu, Dong and Yan, Ning and Li, Houqiang and Wu, Feng}, journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, year={2020}, publisher={IEEE} }
Update Notes
2022.10.28 iWave++ has been accepted as one of the three reference softwares for IEEE 1857.11 Standard for Neural Network-Based Image Coding. Pytorch code (including both training and testing code) and pre-trained models can be found at (https://gitlab.com/iWave/iwave). To further improve compression performance, a series of improvements (such as an enhanced context model and an enhanced de-quant model) have been introduced in the new pytorch implementation compared to the original tensorflow version.
2020.9.4 Upload code and models of Lossy multi-model iWave++.
2020.8.26 Init this repo.
How To Test
-
Dependencies. We test with MIT deepo docker image.
-
Clone this github repo.
-
Place Test images. (The code now only supports images whose border length is a multiple of 16. However, it is very simple to support arbitrary boundary lengths by padding.)
-
Download models. See model folder.
-
python main_testRGB.py. (The path in main_testRGB.py needs to be modified. Please refer to the code.)
Results
iWave++ outperforms Joint, Variational, and iWave. For more information, please refer to the paper.
- RGB PSNR on Kodak dataset.
- RGB PSNR on Tecnick dataset.
